Opportunistic mobile social networks: From mobility and Facebook friendships to structural analysis of user social behavior
Introduction
The vision of a near future in which a multitude of human-driven mobile devices can easily create local wireless networks outside the public Internet is increasingly attracting several groups of researchers in the areas of Delay Tolerant Networks (DTNs) [13], [24], [44], opportunistic networks [39] and the more recent Do-it-yourself (DIY) networks [2]. Considering the wide diffusion of these mobile devices (e.g., smartphones, tablets, etc.) and the impact their use has in the social life of every individual, the study of infastructureless networks allowing short-range (e.g., Bluetooth and Wi-Fi) wireless communication between nodes is generating a particularly hot research trend. When there is no suitable network architecture like the Internet one, for example, an alternative option for communication is necessary. DTNs were designed to allow communication between devices distributed within a networking scenario without fixed network infrastructure, forming sparse network topologies and having intermittent contacts. Using the store-carry-forward communication paradigm, the mobile DTN node first stores the message, then carries it while moving, and then forwards it to an intermediate node or to the destination. A similar strategy is used in opportunistic networks where mobile hand-held nodes forward messages during an encounter opportunity. However, while in DTNs there are also cases where the points of disconnections are known and routing can be performed in an Internet-like fashion, opportunistic networks routes are always computed dynamically.
Opportunistic networks have been shown to be the suitable architecture for several applications in scenarios where network coverage is poor (e.g., dead spots, disaster-recovery situations, etc.) or network access is expensive. Through an opportunistic network and cooperative sensing, for example, it is possible to build sensing maps of air quality, noise, temperature, CO2 concentration, etc., satisfying a specific sensing quality with low delay and energy consumption [50]. Another interesting opportunistic networking application refers to recommender systems. Such solution tracks users’ activities and mobility patterns, and utilizes the user’s contextual information to provide recommendations on a variety of items [33]. Opportunistic networks are also used for mobile data offloading in order to reduce the load on 3G networks [31]: with the increasing number of smartphone users, in fact, most of the 3G networks have been shown to be often overloaded. Another well-known application proposed for such networks is MobiClique [41]: a mobile social networking middleware exploiting ad hoc social networks to disseminate content and leveraging existing social networks to bootstrap the system. More recently, opportunistic networks have been proposed as a promising technology also for big data computing [48] and for connecting smart things in IoT exploiting the social side of things linked to human mobility [30].
It is a common belief that opportunistic networks are characterized by a social-based nature that can be exploited to exchange information. Recent works on opportunistic social routing have shown that message delivery can be optimized selecting the best relay nodes considering both their real-world (i.e., offline) wireless encounters and their online social interactions with the other nodes [5], [15], [16], [21], [36], [41], [45], [47]. Similarly to wireless encounters from which extracting the offline social behavior of nodes, online social networking services like Facebook, Twitter and LinkedIn, just to provide some examples, are fostering the availability of additional data useful for analyzing the overall social behavior of the opportunistic network nodes. From this perspective, we believe that the analysis of sociality derived both from wireless encounters and online data becomes a fundamental aspect within these networks.
There are a lot of works studying encounters of people and social relations. However, while many of these works focus on analyzing the offline sociality extracted from encounters self-reported or detected from wireless proximity, or online social relations such as Facebook friendships or Twitter interactions, few of them analyze both aspects. Moreover, the few works analyzing both offline and online sociality in order to understand how a user behaves within the two contexts, often rely on self-reported meetings that differently from detected Wi-Fi or Bluetooth contacts, may be erroneous because a user may not recall meetings correctly or decide to provide wrong information. This work, differently from the aforementioned works, explicitly focuses on analyzing the relationship between offline sociality built on wireless encounters and online sociality, where both socialities are built on the same set of users.
In these last years, several real-world mobility traces for opportunistic network environments have been collected to explore human-driven motion and sociality. These traces, including in some cases nodes’ online social data/profiles, are usually acquired through experiments tracking a set of participants carrying small portable wireless devices in campuses, conferences, entertainment environments, etc. Most of these data can be obtained through the CRAWDAD1 repository. Although the work done so far analyzing these traces evinced many important aspects on mobility data, the relationship between the sociality built on mobility and the other social dimensions has not been fully discovered yet. In our view, the knowledge about the whole social behavior of mobile users is essential for designing effective social-based algorithms for opportunistic networks. As such, the core of the analysis proposed in this work is represented by the use of a multi-layer network approach for comparing different kinds of social network layers extracted by an heterogeneous set of six mobility traces covering several networking environments: academic, conference and urban scenarios.
Sociologists, anthropologists and psychologists have largely studied human behavior using two different approaches. One approach, being egocentric, focuses on the individual, taking into account his personal network composed by the other individuals to which is directly connected. The other approach, being sociocentric, focuses on large groups of individuals, quantifying internal relations and highlighting any interaction pattern that influences group dynamics. The aim of this work is to study opportunistic nodes’ social behavior using both sociocentric and egocentric network measures [46]. Specifically, we present a detailed analysis of six datasets for mobile social opportunistic networks containing two layers of sociality: the social network graph built on offline wireless encounters and the online social network graph built on Facebook friendships. Exploiting a multi-layer social network approach, we aim to contribute to enlarge the knowledge about the similarity between online and offline worlds in different opportunistic networking environments. This is a much more advanced analysis compared to other recent studies such as [43], [46]. Firstly, we consider a representative collection of six different datasets thus extracting more meaningful conclusions with respect to one single dataset. Secondly, we propose a novel analysis methodology based on egocentric and sociocentric measures for examining the datasets considered. Finally, we define several similarity layers that will be used to uniform the results obtained through the different analysis approaches thus making easier an overall comparison between social network layers.
As a preliminary step, we focus on node centrality (i.e., the contribution of network position to the importance of an individual in the network), thus answering to the challenging question whether online and offline node centralities are correlated and hence, the two social behaviors are similar. Later, we focus on communities, analyzing the similarity between online and offline groups. Starting from this analysis, we exploit the communities detected for investigating online and offline brokerage roles (i.e., nodes that act as brokers between communities) and perform a correlation analysis between online and offline brokerage values. Finally, motivated by recent studies [15], [17], [35], [43] demonstrating that mobile nodes encounter other online socially connected nodes with high probability, we compute offline tie strength in order to find matchings between strong ties and Facebook friendships.
The paper has been organized as follows. Section 2 provides background information on multi-layer social network models and analysis. Section 3 describes the datasets analyzed. Section 4 briefly details the social network model adopted. Section 5 and 6 describe the sociocentric and the egocentric measures used to perform our analysis. Finally, in Section 7, we present our results, provide an overall comparison between the analysis methods used in Section 8 and draw the main conclusions in Section 9.
Section snippets
Related works
The relationship between human encounters and online social relations has been the focus of several researches in these last years. In [32], for example, Hossman et al. analyze two datasets of self-reported data about social, mobility and communication ties of online social network users (Facebook, Twitter and Gowalla) showing that social ties are tightly coupled with mobility and also with communication. Dunbar et al., [23] explore the layered structure of the nodes within two Facebook
Datasets description
To analyze online and offline sociocentric and egocentric behaviors of mobile opportunistic nodes, we consider a collection of six real-world datasets including mobility data and online social data. Most of these datasets can be accessed on the CRAWDAD archive and will be shortly described in this section.
Table 1 summarizes the characteristics of the selected datasets in terms of wireless contacts data. For each experimental dataset, the group of researchers who carried out the experiments
Multi-layer social network model
Starting from mobility and Facebook data, we construct a multi-layer social network graph for each dataset and then, we use this structure to perform sociocentric and egocentric analysis. In this paper, we define a multi-layer social network as in [34], and consider unweighted graph layers since we have Facebook links (friendships) without weights.
- •
Definition 1 (Social Network Layer). A social network layer L is an unweighted graph G(V, E) with vertex set V corresponding to users on the social
Sociocentric analysis
Sociocentric analysis method extends and complements traditional social science by focusing on the quantification of interactions among a socially well-defined group of people and the identification of global structural patterns. In particular, the method analyzes sets of relationships among nodes that are considered as bounded social collectives. In this section, we briefly describe the sociocentric centrality measures and the community detection algorithms that we have chosen within this work
Egocentric analysis
In the previous section, we focused on groups describing some measures that we will use to analyze user sociocentric behaviors in the DSN and the OSN. In order to understand the similarities between the two network layers focusing on the local behavior of individuals more locally, we need to take a closer look to their local circumstances. Egocentric networks are defined as networks of single actors together with the actors they are directly connected to. In this section, we focus on egocentric
Centrality correlation
We initially show the results obtained by computing sociocentric and egocentric centrality measures on the multi-layer networks built on the datasets presented in Section 3. Specifically, within each multi-layer network and for each centrality measure considered, we computed the Pearson’s correlation coefficient between the centrality values of the nodes on the OSN and their centrality values on the DSN. The Pearson’s correlation coefficient is defined as where COV(X, Y) is
Discussion
By proposing a novel methodology for analyzing data based on egocentric and sociocentric measures, we categorized some well-known measures in distinct approaches in order to have a broader view of the user social behavior. Previous works only exploited a subset of the measures we have chosen thus providing partial results. In this section, we summarize the result of the different analyses. This will allow the reader to have an overall clear idea of the social behavior of mobile opportunistic
Conclusions and future work
In this paper, we have presented a novel and detailed methodology for analyzing a set of real mobility traces for opportunistic networks using a multi-layer network approach. The aim of this study has been to better understand user social behavior in terms of egocentric and sociocentric behaviors that can be derived not only from mobility data (encounters’ social network) but also from the available additional information provided by the social network layer built on Facebook friendships. Our
Acknowledgement
We would like to thank the area editor and the anonymous reviewers for their insightful suggestions.
References (50)
- et al.
Egocentric online social networks: Analysis of key features and prediction of tie strength in facebook
Comput. Commun.
(2013) - et al.
Interest-awareness in data dissemination for opportunistic networks
Ad Hoc Netw.
(2015) Residual closeness in networks
Physica A
(2006)- et al.
The structure of online social networks mirrors those in the offline world
Soc. Netw.
(2015) Centrality in social networks conceptual clarification
Social Networks
(1978)- et al.
Collection and analysis of multi-dimensional network data for opportunistic networking research
Comput. Commun.
(2012) - et al.
ML-SOR: Message routing using multi-layer social networks in opportunistic communications
Comput. Netw.
(2015) - ...
- et al.
Do it yourself networking: an interdisciplinary approach
Dagstuhl reports
(2014) - et al.
Foundations of multidimensional network analysis
International Conference on Advances in Social Networks Analysis and Mining (ASONAM)
(2011)